Author: Vincent Diepeveen
Date: 18:26:08 01/04/06
Go up one level in this thread
On January 04, 2006 at 08:59:03, Ratko V Tomic wrote: >I don't know if they can test it. The only code which is high enough level would >be permutation coder or radix coders. The general entropy coding stuff is very >low level, just the basic inner loop coder gear (where the essential differences >from the arithmetic coding are), which one needs to combine into a full coder >which serializes entire output into a single package. If they want to run >benchmarks, I would probably need to extract and clean up for public consumption >the code I used for benchmarking in the main research prototype program. Or if >they set up arithmetic coder under the same conditions and time just its inner >coder loops and use the size of its raw index (removing thus the model info, >data sizes or adaptation), they could compare apples with apples. Readme.txt has >some tips on this kind of test. I guess, I'll wait and see what they need or ask >for the testing. Yes they are very interested. That community is very big. There is big nerds who will follow any instruction you give to test any form of compression you wrote. Arithmetic encoding, a very compact way to write things, is of course useless to computerchess because of the huge CPU resources you need to decompress data. So i dropped arithmetic encoding for that reason. Additional modulo and division are the slowest instructions in existance on the processor. Huffman types are far more interesting to EGTBs therefore. Additional that LZMA+PPM2 type compressors just destroy any arithmetic coder i came up with. Not only those destroyed my own encoding, but also Kadatch fast and superior compressor for EGTBs just finishes off anything i had. Yet if you have something that supercompresses in a lossless way very well, or compresses in a lossy way very well, the geeks there will be very interested. They are real good testers and great benchmarkers. Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.